AI in HR without handing over judgement

The problem it solves

A candidate asks why she was ruled out. The question is reasonable, it arrives by email, and the company discovers it has no answer it would be willing to put in writing. There was a filter, there was an ordering, there was a shortlist, and nobody can reconstruct what actually weighed, or when.

The same gap turns up elsewhere in the department. A performance review is written with a care the file does not support. A job description goes out with requirements nobody compared against the actual work. The contents of a full personnel file, with health details, family details, pay and notes from a private conversation, get pasted into an outside service because a quick summary was needed.

In this area the cost of not being able to explain a decision is not administrative. A decision about a person that cannot be traced back to a written criterion is, in practice, a decision the organization does not control: it cannot defend it, cannot repeat it consistently, and cannot correct it. On the day somebody questions it, the file is all there is.

What it covers

The workshop opens by separating two things that blur together in the working week. On one side sits administrative work, sorting files, preparing draft job descriptions, pulling scattered information into one place, where outside help saves time without touching anyone's judgement. On the other sit decisions about people, where what is at stake is whether the organization can later explain what it did.

The central block is about written criteria. Through a structured self-assessment, each participant revisits one of their own hiring or review processes and checks whether the criterion existed before the decision or was assembled afterwards to justify it. The group then drafts criteria for their own roles against a demanding standard: somebody else should be able to apply them to the same cases and arrive somewhere similar.

The third block covers sensitive information and explainability. Employee and candidate information is classified by whether it may leave the company, with particular attention to what never leaves under any circumstances. In a role play, one participant explains to a member of their team why a decision affecting them was taken, and the room judges whether the explanation holds without mentioning any tool or any system.

The close assigns responsibility. Who reviews at each step, what the team is told about how these tools are used and within what limits, and what remains for the leadership to decide. Participants leave with an action plan for the coming weeks, sorted into what can be done without anyone's permission and what needs a decision from above.

What participants learn

• Identify the tasks where outside help saves administrative effort without touching a decision.
• Establish what employee and candidate information never leaves the company under any circumstances.
• Draft the criterion behind a decision about a person, whether or not a tool is involved.
• Assign human review and a named owner at every step of hiring and performance review.
• Develop the message that tells the team what is used, for what, and within what limits.

Agenda

1. 01Administrative work in HR versus decisions about people
2. 02Sensitive information: what is shared, with whom, and what never is
3. 03Job descriptions, interview guides and first drafts
4. 04Evaluation criteria written before any tool arrives
5. 05Bias, explainability and a person's right to an answer
6. 06Mandatory human review and who answers for the outcome
7. 07What the team is told about how these tools are used

What the organization leaves with

• A written evaluation criterion that predates any tool and applies to the roles in the department.
• A classification of employee and candidate information by whether it may leave the company.
• A human review point with a named owner at each step of the hiring process.
• A drafted message to the team about what is used in the department, for what, and within what limits.